An investigation into the relationship between the Big Five, shyness, narcissism, loneliness, and Facebook usage.
Main focus of the article:
My comments on the article:
The stated aim of this study is to try and identify which personality characteristics provide the best predictors of Facebook usage using their 1324 self-selected participants. The self-selection occurred through advertisements placed on Facebook and in “threads on six popular Australian online discussion forums – Best Recipes, Big Footy, Bub Hub, Essential Baby, MacTalk and VicHorse” (p. 1661). The authors provide no indication of how these particular discussion forums were identified as “popular” and, given that I am in their target demographic, find it concerning that I was unaware of any of these forums prior to reading this article. Equally, given that “Bub Hub” and “Essential Baby” are both forums relating to pregnancy and early childhood, this potentially introduces further bias, both in terms of gender and personality types.
Aside from the obvious potential biases and problems with self-selected samples; the ratio of Facebook users to non-users is not reflective of the stated estimated population proportion at the beginning of the article, viz, “Enthusiasm for Facebook is particularly apparent in Australia, as close to half of the population are reported to be active users (Gonzalez, 2011)” (p. 1658). With approximately 87.5% of participants being Facebook users it suggests that this is potentially a source of bias. The authors state, “In order to obtain a sample that was representative of average Australian Facebook users (Gonzalez, 2011), participants were required to be between 18 and 44 years old”. Checking both the recently released 2011 Australian Census data and today’s data on Facebook usage (assuming that they are comparable) we see that approximately 90% of the targeted age range (18-44) are indicated to be Facebook users and as such, the ratio in this study is, perhaps, reflective of the population of interest, however, the authors do not make this clear. Nor do they make clear if the gender breakdown is also representative.
Ryan and Xenos also report that for both Facebook users and non-users “Mage group = 25–34″ which presumably, despite the APA symbology for mean being used, is meant to indicate that modal age group of their participants. It would have been nice if the authors could have provided at least a histogram of the age distribution, preferably in the same 5-year ranges as the Australian Bureau of Statistics (ABS) use in reporting the census data, to enable comparison to population age group distribution.
The results section of the article provides details of the statistical analyses performed on the collected data. Ryan and Xenos start by indicating that, “Normality was tested for each of the personality measures by assessing stem-and-leaf plots and histograms” (p. 1661) although why they have not used standard tests like the Kolmogorov-Smirnov or Shapiro-Wilk tests is concerning. This concern is further heightened when the authors indicate that because “the sample was large, skewness and kurtosis statistics were not calculated (Field, 2005)” (p. 1661). Despite the large sample size (as per the reference to Field), skewness and kurtosis statistics would help to allay concerns about the shape of the distribution given that standard normality test statistics were not used and graphical representations of the data are not provided.
The first major analysis presented is that of a MANOVA performed on the BFI data. To be clear, in MANOVA the procedure uses one (or more) independent variables with multiple dependent variables. If the results indicate that you should reject the Null Hypothesis, then post-hoc testing is required to determine where potential differences occur. Ryan and Xenos report that their analysis indicates that “Facebook users had significantly higher scores on extraversion” and “significantly lower scores on conscientiousness” compared to non-users (p. 1661). From this we can infer that if a person is a Facebook user, we would expect their extraversion score to be higher than the population mean score for non-users. Similarly, we can infer that if a person is a Facebook user, we would expect their conscientiousness score to be lower than the population mean score for non-users. In other words, we expect that Facebook users are more extroverted and less conscientiousness than non-users.
The second MANOVA is performed on the “shyness and loneliness, as well as the three sub-factors of loneliness: romantic, family, and social loneliness” (p. 1661) data which presumably comes from some combination of the RCBS and SELSA-S data. Ryan and Xenos report that “Facebook nonusers were significantly more likely than users to be shy … and to experience social loneliness” whereas “Facebook users had significantly higher levels of family loneliness” (p. 1661). They also note that the “dependent variables of total loneliness … and romantic loneliness … were not significantly” (p. 1661). Again, this means that if one knows whether or not a person is a Facebook user, an inference can be made about their expected values on these scales where significant results are found.
The final MANOVA is performed on the data on “narcissism, as well as two of its sub-factors – exhibitionism and leadership” (p. 1661) which presumably comes from the NPI-29 data. Ryan and Xenos report that their analysis shows that “Facebook users were significantly more likely than nonusers to have higher levels of total narcissism … exhibitionism … and leadership” (p. 1661). Once again, this means that if one knows whether or not a person is a Facebook user, an inference can be made about their expected values on these scales where significant results are found.
Ryan and Xenos then present an analysis of the frequency of Facebook usage, along with Pearson correlation statistics relating these usage data with personality types, however, no demonstration is made to assure the reader that the linearity assumption has been met for the Pearson correlation to be valid. Equally, all of the presented correlation coefficients are rather small values ( ) which suggests that these statistically significant correlations are unable to explain at least 96% of the observed variability; hardly something one would get excited about. At the same time, these low values for a Pearson’s Correlation should be encouraging the researchers to look more closely at the data to identify if there is really a relationship present in the data and assessing sub groups (e.g. by gender). Similar comments also hold for their analysis of the participant’s preference for Facebook features.
Ryan and Xenos indicate that a principle component factor analysis with Varimax rotation is performed on the data, identifying four factors, namely “Active Social Contributions, Passive Engagement, News and Information, and Real-Time Social Interaction” (p. 1662). Since Varimax is an orthogonal rotation, it assumes that the factors will be unrelated (i.e. independent), however, their selected factor names do not suggest that this is what they believe to be true and hence an oblique rotation (e.g. Oblimin) would have been a better choice. In addition, the choice of a principle component analysis method instead of maximum likelihood is interesting, given that this choice severely restricts the generalisability of their results, something that a close reading of the cited Field text would have highlighted.
The discussion section of this article states that “the prediction that extraverted people would be more likely to use Facebook than introverted people was supported” (p. 1662). This is akin to saying that since “All poodles are dogs” this implies that since I have observed a dog it must therefore be a poodle. To put it in terms of their study, this means that if “All Facebook users are extroverts” then having identified someone as an extrovert you conclude that they are therefore a Facebook user. The logic and mathematics of a MANOVA test are not reversible in this way and as clearly identified earlier in their report, Ryan and Xenos use Facebook user status as an independent variable in their analysis. It cannot be then used as a dependent variable in their conclusions.
The choice of statistical method here is, in my opinion, the main source of the problems with their conclusions. If Ryan and Xenos want to predict Facebook usage based upon identified personality traits, in other words using Facebook usage as the dependent variable, then the most appropriate statistical method would be a form of logistic regression (especially as the dependent variable is dichotomous). The conclusion could then be made about the increased or decreased odds of being a Facebook user (compared to being a non-user) based upon the identified personality traits, gender, age or other independent variables.
The other major problem with the conclusions of this article is that, despite the attempted rigour in their statistical analyses, the sampling technique used is unlikely to result in a truly representative sample of the population of interest. A point acknowledged by the authors, “Secondly, the methods of recruitment employed in this study may have led to sample bias” (p. 1663). As such, any conclusions made are difficult to generalise beyond that sample. In short, the authors would have been better served to make this article into a descriptive piece rather than attempting to make inferences from their sample data.
Full citation and link to document:
Ryan, T., & Xenos, S. (2011). Who uses Facebook? An investigation into the relationship between the Big Five, shyness, narcissism, loneliness, and Facebook usage. Computers in Human Behavior, 27(5), 1658-1664. doi: 10.1016/j.chb.2011.02.004
CITE THIS AS:
Ovens, Matthew. “Ryan & Xenos (2011) Who uses Facebook?” Retrieved from YourStatsGuru.
First published 2012 | Last updated 21 January 2018